OpenAI accused of training Astra on private math conversations to claim breakthrough

OpenAI might have stolen another major proof

Andreas Thom, a leading expert on sofic groups, says in a Mastodon post that OpenAI may have trained its Astra model on private discussions about Gromov's soficity conjecture, one of ten problems OpenAI later announced Astra had solved. Valerio Capraro, who knows Thom, calls the claim credible and warns that if similar allegations by Levent Alpöge and Tristan Buckmaster are substantiated, it could become one of the greatest intellectual scandals in science.